Re: Looking for bottleneck during load test - Mailing list pgsql-performance

From Hell, Robert
Subject Re: Looking for bottleneck during load test
Date
Msg-id B710F3299F04664DB6B37C258FDEEB94015F1DD7@FABAMAIL.fabagl.fabasoft.com
Whole thread Raw
In response to Re: Looking for bottleneck during load test  (Ivan Voras <ivoras@geri.cc.fer.hr>)
List pgsql-performance
Worst latency was at ~600 ms.

In this test case we have only selects - so it's read only.

I tried 8.3 - it's better there (10-15 %) - but IO rates stay the same.
I use 18 GB shared memory on a machine with 32 GB during the test, I think this shouldn't be a problem.

I did some further testing and I now really think that disk is the bottleneck here. 
I'm really surprised that those 7 MB/s in average are the maximum for that disks.

Any filesystem (ext3) settings that could help here?

Regards,
Robert


-----Original Message-----
From: pgsql-performance-owner@postgresql.org [mailto:pgsql-performance-owner@postgresql.org] On Behalf Of Ivan Voras
Sent: Dienstag, 08. April 2008 11:03
To: pgsql-performance@postgresql.org
Subject: Re: [PERFORM] Looking for bottleneck during load test

Hell, Robert wrote:
> > I tried different other tools for random IO (including a self
written one which does random lseek and read).
> >
> > This tool, started during one of our tests, achieves 2 iops (8k each).
> > Started alone I get something about 1,500 iops with an avg latency
of 100 ms.

1500 iops looks about right for 4x2 RAID 10 volume. What's your worst
latency (as reported by the tool)? iowait is mostly seek time.

> > We are using SAN (EMC CLARiiON CX 300) - are those ~7 MB/s really
our bottleneck?

Depending on your access pattern to the database, it could be (if you
have lots of random IO, and 180 concurrent database threads can make any
IO random enough). Are your queries read-mostly or a mix?

> > Any other tuning ideas?

Only generic ones:

- Are your queries optimized, use indexes, etc.?
- Try PostgreSQL 8.3 - if you have sequential seeks it can in theory
make better use of data between connections.
- Do you have enough memory dedicated to data caches, both in PostgreSQL
and in the OS? (i.e. what is your shared_buffers setting?)
- If the SAN can configure parameters such as prefetch (pre-read) and
stripe size, try lowering them (should help if you have random IO).


-- 
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance

pgsql-performance by date:

Previous
From: Ivan Voras
Date:
Subject: Re: Looking for bottleneck during load test
Next
From: Matthew
Date:
Subject: Re: Forcing more agressive index scans for BITMAP AND